filesystem-context

You can persist agent context to disk by offloading large tool outputs and scratch notes to files, allowing long-running multi-turn agent workflows to access past memory without bloating the active prompt.

Sharing state across sub-agents is best handled by offloading data to a filesystem, enabling coordinated sub-agent workspaces and dynamic context loading for memory-efficient workflows.

You can offload large tool outputs to scratch files using a tool output handler, referencing these files in context later to prevent active prompt bloat during multi-turn tasks.

Yes, you can persist agent plans and scratch pads across sessions by writing them to the filesystem, ensuring learned preferences and task states remain accessible for future runs.

Filesystem-based context management is needed for long-running, multi-turn agent workflows that require plan persistence, tool-output offloading, and state sharing across sub-agents.

Yes, the approach supports dynamic loading of skills by maintaining per-skill directories on disk, allowing agents to load required components and references on demand.